Dario Amodei, CEO of Anthropic, recently shared insights on the progress of Chinese AI enterprise, DeepSeek, arguing that its achievements demonstrate the efficacy of US AI export controls.
Previously, Amodei, together with ex-deputy national security advisor, Matt Pottinger, advocated tighter export controls. Now, in his article, he maintains that the current regulations continue to put a check on Chinese AI companies like DeepSeek.
He highlighted that, given the release time frame, DeepSeek’s AI models underperform when compared with their American counterparts. “DeepSeek’s output is reflective of US models from 7-10 months ago, albeit at lower costs. It’s simply an anticipated point on the ongoing cost reduction curve,” defends Amodei.
DeepSeek’s flagship model, DeepSeek V3, was compared to Anthropic’s Claude 3.5 Sonnet. Despite being trained months later, Sonnet outclasses DeepSeek V3 in various evaluations.
Amodei denotes U.S. companies’ notable strides in cost reduction, claiming DeepSeek’s efficiency innovations will soon be implemented by both US and Chinese labs.
Amodei warns that the future trajectory hinges upon established export controls. He speculates that stricter export rules can enable the US and its allies to gain a significant and lasting advantage. Conversely, if China is allowed easier access to crucial AI chips, Amodei fears China may leverage AI for military applications.
Howard Lutnick, Trump’s choice for the commerce secretary, has echoed these sentiments, accusing DeepSeek of intellectual property theft and advocating for more robust tariffs to support controls.
Anthropic’s rival, OpenAI, also urges more aggressive measures to safeguard US dominance in AI, suggesting an influx of global funds into AI projects to counteract potential Chinese influence.
Original source: Read the full article on TechCrunch
